Turney & Littman's (2003) Semantic Orientation from Association

This is an implementation of the semantic/sentiment similarity method developed by Turney and Littman 2003. The underlying matrix is a word × word matrix extracted from the NYT section of English Gigaword. The rows are the Harvard Inquirer words and the columns are the full vocabulary. (Both rows and columns are restricted to items with at least 60 tokens in the data.) The notion of co-ocurrence is a restricted one: sharing a semantic dependency in a Stanford collapsed dependency representation. PPMI with contextual discounting was applied to the count matrix, truncated SVD was applied to that (150 dimensions), and then pairwise cosine similarity was defined for the row (Inquirer) vocabulary.

Provide two seed-sets drawn from the Harvard Inquirer vocabulary (comma-separated string)

Seed set 1:

Seed set 2:

or select one of the following random five-word subsets of some Harvard Inquirer oppositions:

Turney–Littman Pos: good, nice, excellent, positive, fortunate, correct, superior
Turney–Littman Neg: bad, nasty, poor, negative, unfortunate, wrong, inferior
Positiv: mighty, dignified, wonder, thank, brainy
Negativ: obsolete, ghastly, anarchist, heinous, false
Strong: prohibit, exploit, bullet, consistent, domination
Weak: fatigue, loser, break, sucker, serve
Active: produce, sign, derive, demean, surveillance
Passive: croak, project, realize, sprawl, fatigue
Pleasur: love, uplift, willing, attraction, cheery
Pain: outrage, cramp, upset, malice, displeasure
Ovrst: astronomical, impressive, most, prompt, assurance
Undrst: large, well, unsure, event, temperance
Yes: agree, hurrah, yeah, course, okay
No: disagree, wrong, nay, nope, mean